xref: /freebsd/usr.sbin/pmcstat/pmcstat.c (revision b00fe64f4acfe315181f65999af16e9a7bdc600b)
1 /*-
2  * Copyright (c) 2003-2008, Joseph Koshy
3  * Copyright (c) 2007 The FreeBSD Foundation
4  * All rights reserved.
5  *
6  * Portions of this software were developed by A. Joseph Koshy under
7  * sponsorship from the FreeBSD Foundation and Google, Inc.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/cpuset.h>
36 #include <sys/event.h>
37 #include <sys/queue.h>
38 #include <sys/socket.h>
39 #include <sys/stat.h>
40 #include <sys/sysctl.h>
41 #include <sys/time.h>
42 #include <sys/ttycom.h>
43 #include <sys/user.h>
44 #include <sys/wait.h>
45 
46 #include <assert.h>
47 #include <curses.h>
48 #include <err.h>
49 #include <errno.h>
50 #include <fcntl.h>
51 #include <kvm.h>
52 #include <libgen.h>
53 #include <limits.h>
54 #include <math.h>
55 #include <pmc.h>
56 #include <pmclog.h>
57 #include <regex.h>
58 #include <signal.h>
59 #include <stdarg.h>
60 #include <stdint.h>
61 #include <stdio.h>
62 #include <stdlib.h>
63 #include <string.h>
64 #include <sysexits.h>
65 #include <unistd.h>
66 
67 #include "pmcstat.h"
68 
69 /*
70  * A given invocation of pmcstat(8) can manage multiple PMCs of both
71  * the system-wide and per-process variety.  Each of these could be in
72  * 'counting mode' or in 'sampling mode'.
73  *
74  * For 'counting mode' PMCs, pmcstat(8) will periodically issue a
75  * pmc_read() at the configured time interval and print out the value
76  * of the requested PMCs.
77  *
78  * For 'sampling mode' PMCs it can log to a file for offline analysis,
79  * or can analyse sampling data "on the fly", either by converting
80  * samples to printed textual form or by creating gprof(1) compatible
81  * profiles, one per program executed.  When creating gprof(1)
82  * profiles it can optionally merge entries from multiple processes
83  * for a given executable into a single profile file.
84  *
85  * pmcstat(8) can also execute a command line and attach PMCs to the
86  * resulting child process.  The protocol used is as follows:
87  *
88  * - parent creates a socketpair for two way communication and
89  *   fork()s.
90  * - subsequently:
91  *
92  *   /Parent/				/Child/
93  *
94  *   - Wait for childs token.
95  *					- Sends token.
96  *					- Awaits signal to start.
97  *  - Attaches PMCs to the child's pid
98  *    and starts them. Sets up
99  *    monitoring for the child.
100  *  - Signals child to start.
101  *					- Receives signal, attempts exec().
102  *
103  * After this point normal processing can happen.
104  */
105 
106 /* Globals */
107 
108 int		pmcstat_displayheight = DEFAULT_DISPLAY_HEIGHT;
109 int		pmcstat_displaywidth  = DEFAULT_DISPLAY_WIDTH;
110 static int	pmcstat_sockpair[NSOCKPAIRFD];
111 static int	pmcstat_kq;
112 static kvm_t	*pmcstat_kvm;
113 static struct kinfo_proc *pmcstat_plist;
114 struct pmcstat_args args;
115 
116 static void
117 pmcstat_clone_event_descriptor(struct pmcstat_ev *ev, const cpuset_t *cpumask)
118 {
119 	int cpu, mcpu;
120 	struct pmcstat_ev *ev_clone;
121 
122 	mcpu = sizeof(*cpumask) * NBBY;
123 	for (cpu = 0; cpu < mcpu; cpu++) {
124 		if (!CPU_ISSET(cpu, cpumask))
125 			continue;
126 
127 		if ((ev_clone = malloc(sizeof(*ev_clone))) == NULL)
128 			errx(EX_SOFTWARE, "ERROR: Out of memory");
129 		(void) memset(ev_clone, 0, sizeof(*ev_clone));
130 
131 		ev_clone->ev_count = ev->ev_count;
132 		ev_clone->ev_cpu   = cpu;
133 		ev_clone->ev_cumulative = ev->ev_cumulative;
134 		ev_clone->ev_flags = ev->ev_flags;
135 		ev_clone->ev_mode  = ev->ev_mode;
136 		ev_clone->ev_name  = strdup(ev->ev_name);
137 		ev_clone->ev_pmcid = ev->ev_pmcid;
138 		ev_clone->ev_saved = ev->ev_saved;
139 		ev_clone->ev_spec  = strdup(ev->ev_spec);
140 
141 		STAILQ_INSERT_TAIL(&args.pa_events, ev_clone, ev_next);
142 	}
143 }
144 
145 static void
146 pmcstat_get_cpumask(const char *cpuspec, cpuset_t *cpumask)
147 {
148 	int cpu;
149 	const char *s;
150 	char *end;
151 
152 	CPU_ZERO(cpumask);
153 	s = cpuspec;
154 
155 	do {
156 		cpu = strtol(s, &end, 0);
157 		if (cpu < 0 || end == s)
158 			errx(EX_USAGE,
159 			    "ERROR: Illegal CPU specification \"%s\".",
160 			    cpuspec);
161 		CPU_SET(cpu, cpumask);
162 		s = end + strspn(end, ", \t");
163 	} while (*s);
164 }
165 
166 void
167 pmcstat_attach_pmcs(void)
168 {
169 	struct pmcstat_ev *ev;
170 	struct pmcstat_target *pt;
171 	int count;
172 
173 	/* Attach all process PMCs to target processes. */
174 	count = 0;
175 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
176 		if (PMC_IS_SYSTEM_MODE(ev->ev_mode))
177 			continue;
178 		SLIST_FOREACH(pt, &args.pa_targets, pt_next)
179 			if (pmc_attach(ev->ev_pmcid, pt->pt_pid) == 0)
180 				count++;
181 			else if (errno != ESRCH)
182 				err(EX_OSERR,
183 "ERROR: cannot attach pmc \"%s\" to process %d",
184 				    ev->ev_name, (int)pt->pt_pid);
185 	}
186 
187 	if (count == 0)
188 		errx(EX_DATAERR, "ERROR: No processes were attached to.");
189 }
190 
191 
192 void
193 pmcstat_cleanup(void)
194 {
195 	struct pmcstat_ev *ev, *tmp;
196 
197 	/* release allocated PMCs. */
198 	STAILQ_FOREACH_SAFE(ev, &args.pa_events, ev_next, tmp)
199 	    if (ev->ev_pmcid != PMC_ID_INVALID) {
200 		if (pmc_stop(ev->ev_pmcid) < 0)
201 			err(EX_OSERR, "ERROR: cannot stop pmc 0x%x \"%s\"",
202 			    ev->ev_pmcid, ev->ev_name);
203 		if (pmc_release(ev->ev_pmcid) < 0)
204 			err(EX_OSERR, "ERROR: cannot release pmc 0x%x \"%s\"",
205 			    ev->ev_pmcid, ev->ev_name);
206 		free(ev->ev_name);
207 		free(ev->ev_spec);
208 		STAILQ_REMOVE(&args.pa_events, ev, pmcstat_ev, ev_next);
209 		free(ev);
210 	    }
211 
212 	/* de-configure the log file if present. */
213 	if (args.pa_flags & (FLAG_HAS_PIPE | FLAG_HAS_OUTPUT_LOGFILE))
214 		(void) pmc_configure_logfile(-1);
215 
216 	if (args.pa_logparser) {
217 		pmclog_close(args.pa_logparser);
218 		args.pa_logparser = NULL;
219 	}
220 
221 	pmcstat_shutdown_logging();
222 }
223 
224 void
225 pmcstat_create_process(void)
226 {
227 	char token;
228 	pid_t pid;
229 	struct kevent kev;
230 	struct pmcstat_target *pt;
231 
232 	if (socketpair(AF_UNIX, SOCK_STREAM, 0, pmcstat_sockpair) < 0)
233 		err(EX_OSERR, "ERROR: cannot create socket pair");
234 
235 	switch (pid = fork()) {
236 	case -1:
237 		err(EX_OSERR, "ERROR: cannot fork");
238 		/*NOTREACHED*/
239 
240 	case 0:		/* child */
241 		(void) close(pmcstat_sockpair[PARENTSOCKET]);
242 
243 		/* Write a token to tell our parent we've started executing. */
244 		if (write(pmcstat_sockpair[CHILDSOCKET], "+", 1) != 1)
245 			err(EX_OSERR, "ERROR (child): cannot write token");
246 
247 		/* Wait for our parent to signal us to start. */
248 		if (read(pmcstat_sockpair[CHILDSOCKET], &token, 1) < 0)
249 			err(EX_OSERR, "ERROR (child): cannot read token");
250 		(void) close(pmcstat_sockpair[CHILDSOCKET]);
251 
252 		/* exec() the program requested */
253 		execvp(*args.pa_argv, args.pa_argv);
254 		/* and if that fails, notify the parent */
255 		kill(getppid(), SIGCHLD);
256 		err(EX_OSERR, "ERROR: execvp \"%s\" failed", *args.pa_argv);
257 		/*NOTREACHED*/
258 
259 	default:	/* parent */
260 		(void) close(pmcstat_sockpair[CHILDSOCKET]);
261 		break;
262 	}
263 
264 	/* Ask to be notified via a kevent when the target process exits. */
265 	EV_SET(&kev, pid, EVFILT_PROC, EV_ADD|EV_ONESHOT, NOTE_EXIT, 0,
266 	    NULL);
267 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
268 		err(EX_OSERR, "ERROR: cannot monitor child process %d", pid);
269 
270 	if ((pt = malloc(sizeof(*pt))) == NULL)
271 		errx(EX_SOFTWARE, "ERROR: Out of memory.");
272 
273 	pt->pt_pid = pid;
274 	SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
275 
276 	/* Wait for the child to signal that its ready to go. */
277 	if (read(pmcstat_sockpair[PARENTSOCKET], &token, 1) < 0)
278 		err(EX_OSERR, "ERROR (parent): cannot read token");
279 
280 	return;
281 }
282 
283 void
284 pmcstat_find_targets(const char *spec)
285 {
286 	int n, nproc, pid, rv;
287 	struct pmcstat_target *pt;
288 	char errbuf[_POSIX2_LINE_MAX], *end;
289 	static struct kinfo_proc *kp;
290 	regex_t reg;
291 	regmatch_t regmatch;
292 
293 	/* First check if we've been given a process id. */
294       	pid = strtol(spec, &end, 0);
295 	if (end != spec && pid >= 0) {
296 		if ((pt = malloc(sizeof(*pt))) == NULL)
297 			goto outofmemory;
298 		pt->pt_pid = pid;
299 		SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
300 		return;
301 	}
302 
303 	/* Otherwise treat arg as a regular expression naming processes. */
304 	if (pmcstat_kvm == NULL) {
305 		if ((pmcstat_kvm = kvm_openfiles(NULL, "/dev/null", NULL, 0,
306 		    errbuf)) == NULL)
307 			err(EX_OSERR, "ERROR: Cannot open kernel \"%s\"",
308 			    errbuf);
309 		if ((pmcstat_plist = kvm_getprocs(pmcstat_kvm, KERN_PROC_PROC,
310 		    0, &nproc)) == NULL)
311 			err(EX_OSERR, "ERROR: Cannot get process list: %s",
312 			    kvm_geterr(pmcstat_kvm));
313 	} else
314 		nproc = 0;
315 
316 	if ((rv = regcomp(&reg, spec, REG_EXTENDED|REG_NOSUB)) != 0) {
317 		regerror(rv, &reg, errbuf, sizeof(errbuf));
318 		err(EX_DATAERR, "ERROR: Failed to compile regex \"%s\": %s",
319 		    spec, errbuf);
320 	}
321 
322 	for (n = 0, kp = pmcstat_plist; n < nproc; n++, kp++) {
323 		if ((rv = regexec(&reg, kp->ki_comm, 1, &regmatch, 0)) == 0) {
324 			if ((pt = malloc(sizeof(*pt))) == NULL)
325 				goto outofmemory;
326 			pt->pt_pid = kp->ki_pid;
327 			SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
328 		} else if (rv != REG_NOMATCH) {
329 			regerror(rv, &reg, errbuf, sizeof(errbuf));
330 			errx(EX_SOFTWARE, "ERROR: Regex evalation failed: %s",
331 			    errbuf);
332 		}
333 	}
334 
335 	regfree(&reg);
336 
337 	return;
338 
339  outofmemory:
340 	errx(EX_SOFTWARE, "Out of memory.");
341 	/*NOTREACHED*/
342 }
343 
344 void
345 pmcstat_kill_process(void)
346 {
347 	struct pmcstat_target *pt;
348 
349 	assert(args.pa_flags & FLAG_HAS_COMMANDLINE);
350 
351 	/*
352 	 * If a command line was specified, it would be the very first
353 	 * in the list, before any other processes specified by -t.
354 	 */
355 	pt = SLIST_FIRST(&args.pa_targets);
356 	assert(pt != NULL);
357 
358 	if (kill(pt->pt_pid, SIGINT) != 0)
359 		err(EX_OSERR, "ERROR: cannot signal child process");
360 }
361 
362 void
363 pmcstat_start_pmcs(void)
364 {
365 	struct pmcstat_ev *ev;
366 
367 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
368 
369 	    assert(ev->ev_pmcid != PMC_ID_INVALID);
370 
371 	    if (pmc_start(ev->ev_pmcid) < 0) {
372 	        warn("ERROR: Cannot start pmc 0x%x \"%s\"",
373 		    ev->ev_pmcid, ev->ev_name);
374 		pmcstat_cleanup();
375 		exit(EX_OSERR);
376 	    }
377 	}
378 
379 }
380 
381 void
382 pmcstat_print_headers(void)
383 {
384 	struct pmcstat_ev *ev;
385 	int c, w;
386 
387 	(void) fprintf(args.pa_printfile, PRINT_HEADER_PREFIX);
388 
389 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
390 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
391 			continue;
392 
393 		c = PMC_IS_SYSTEM_MODE(ev->ev_mode) ? 's' : 'p';
394 
395 		if (ev->ev_fieldskip != 0)
396 			(void) fprintf(args.pa_printfile, "%*s",
397 			    ev->ev_fieldskip, "");
398 		w = ev->ev_fieldwidth - ev->ev_fieldskip - 2;
399 
400 		if (c == 's')
401 			(void) fprintf(args.pa_printfile, "s/%02d/%-*s ",
402 			    ev->ev_cpu, w-3, ev->ev_name);
403 		else
404 			(void) fprintf(args.pa_printfile, "p/%*s ", w,
405 			    ev->ev_name);
406 	}
407 
408 	(void) fflush(args.pa_printfile);
409 }
410 
411 void
412 pmcstat_print_counters(void)
413 {
414 	int extra_width;
415 	struct pmcstat_ev *ev;
416 	pmc_value_t value;
417 
418 	extra_width = sizeof(PRINT_HEADER_PREFIX) - 1;
419 
420 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
421 
422 		/* skip sampling mode counters */
423 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
424 			continue;
425 
426 		if (pmc_read(ev->ev_pmcid, &value) < 0)
427 			err(EX_OSERR, "ERROR: Cannot read pmc \"%s\"",
428 			    ev->ev_name);
429 
430 		(void) fprintf(args.pa_printfile, "%*ju ",
431 		    ev->ev_fieldwidth + extra_width,
432 		    (uintmax_t) ev->ev_cumulative ? value :
433 		    (value - ev->ev_saved));
434 
435 		if (ev->ev_cumulative == 0)
436 			ev->ev_saved = value;
437 		extra_width = 0;
438 	}
439 
440 	(void) fflush(args.pa_printfile);
441 }
442 
443 /*
444  * Print output
445  */
446 
447 void
448 pmcstat_print_pmcs(void)
449 {
450 	static int linecount = 0;
451 
452 	/* check if we need to print a header line */
453 	if (++linecount > pmcstat_displayheight) {
454 		(void) fprintf(args.pa_printfile, "\n");
455 		linecount = 1;
456 	}
457 	if (linecount == 1)
458 		pmcstat_print_headers();
459 	(void) fprintf(args.pa_printfile, "\n");
460 
461 	pmcstat_print_counters();
462 
463 	return;
464 }
465 
466 /*
467  * Do process profiling
468  *
469  * If a pid was specified, attach each allocated PMC to the target
470  * process.  Otherwise, fork a child and attach the PMCs to the child,
471  * and have the child exec() the target program.
472  */
473 
474 void
475 pmcstat_start_process(void)
476 {
477 	/* Signal the child to proceed. */
478 	if (write(pmcstat_sockpair[PARENTSOCKET], "!", 1) != 1)
479 		err(EX_OSERR, "ERROR (parent): write of token failed");
480 
481 	(void) close(pmcstat_sockpair[PARENTSOCKET]);
482 }
483 
484 void
485 pmcstat_show_usage(void)
486 {
487 	errx(EX_USAGE,
488 	    "[options] [commandline]\n"
489 	    "\t Measure process and/or system performance using hardware\n"
490 	    "\t performance monitoring counters.\n"
491 	    "\t Options include:\n"
492 	    "\t -C\t\t (toggle) show cumulative counts\n"
493 	    "\t -D path\t create profiles in directory \"path\"\n"
494 	    "\t -E\t\t (toggle) show counts at process exit\n"
495 	    "\t -F file\t write a system-wide callgraph (Kcachegrind format)"
496 		" to \"file\"\n"
497 	    "\t -G file\t write a system-wide callgraph to \"file\"\n"
498 	    "\t -M file\t print executable/gmon file map to \"file\"\n"
499 	    "\t -N\t\t (toggle) capture callchains\n"
500 	    "\t -O file\t send log output to \"file\"\n"
501 	    "\t -P spec\t allocate a process-private sampling PMC\n"
502 	    "\t -R file\t read events from \"file\"\n"
503 	    "\t -S spec\t allocate a system-wide sampling PMC\n"
504 	    "\t -T\t\t start in top mode\n"
505 	    "\t -W\t\t (toggle) show counts per context switch\n"
506 	    "\t -a file\t print sampled PCs and callgraph to \"file\"\n"
507 	    "\t -c cpu-list\t set cpus for subsequent system-wide PMCs\n"
508 	    "\t -d\t\t (toggle) track descendants\n"
509 	    "\t -f spec\t pass \"spec\" to as plugin option\n"
510 	    "\t -g\t\t produce gprof(1) compatible profiles\n"
511 	    "\t -k dir\t\t set the path to the kernel\n"
512 	    "\t -l secs\t set duration time\n"
513 	    "\t -m file\t print sampled PCs to \"file\"\n"
514 	    "\t -n rate\t set sampling rate\n"
515 	    "\t -o file\t send print output to \"file\"\n"
516 	    "\t -p spec\t allocate a process-private counting PMC\n"
517 	    "\t -q\t\t suppress verbosity\n"
518 	    "\t -r fsroot\t specify FS root directory\n"
519 	    "\t -s spec\t allocate a system-wide counting PMC\n"
520 	    "\t -t process-spec attach to running processes matching "
521 		"\"process-spec\"\n"
522 	    "\t -v\t\t increase verbosity\n"
523 	    "\t -w secs\t set printing time interval\n"
524 	    "\t -z depth\t limit callchain display depth"
525 	);
526 }
527 
528 /*
529  * At exit handler for top mode
530  */
531 
532 void
533 pmcstat_topexit(void)
534 {
535 	if (!args.pa_toptty)
536 		return;
537 
538 	/*
539 	 * Shutdown ncurses.
540 	 */
541 	clrtoeol();
542 	refresh();
543 	endwin();
544 }
545 
546 /*
547  * Main
548  */
549 
550 int
551 main(int argc, char **argv)
552 {
553 	cpuset_t cpumask;
554 	double interval;
555 	double duration;
556 	int hcpu, option, npmc, ncpu;
557 	int c, check_driver_stats, current_sampling_count;
558 	int do_callchain, do_descendants, do_logproccsw, do_logprocexit;
559 	int do_print, do_read;
560 	size_t len;
561 	int graphdepth;
562 	int pipefd[2], rfd;
563 	int use_cumulative_counts;
564 	short cf, cb;
565 	char *end, *tmp;
566 	const char *errmsg, *graphfilename;
567 	enum pmcstat_state runstate;
568 	struct pmc_driverstats ds_start, ds_end;
569 	struct pmcstat_ev *ev;
570 	struct sigaction sa;
571 	struct kevent kev;
572 	struct winsize ws;
573 	struct stat sb;
574 	char buffer[PATH_MAX];
575 
576 	check_driver_stats      = 0;
577 	current_sampling_count  = DEFAULT_SAMPLE_COUNT;
578 	do_callchain		= 1;
579 	do_descendants          = 0;
580 	do_logproccsw           = 0;
581 	do_logprocexit          = 0;
582 	use_cumulative_counts   = 0;
583 	graphfilename		= "-";
584 	args.pa_required	= 0;
585 	args.pa_flags		= 0;
586 	args.pa_verbosity	= 1;
587 	args.pa_logfd		= -1;
588 	args.pa_fsroot		= "";
589 	args.pa_samplesdir	= ".";
590 	args.pa_printfile	= stderr;
591 	args.pa_graphdepth	= DEFAULT_CALLGRAPH_DEPTH;
592 	args.pa_graphfile	= NULL;
593 	args.pa_interval	= DEFAULT_WAIT_INTERVAL;
594 	args.pa_mapfilename	= NULL;
595 	args.pa_inputpath	= NULL;
596 	args.pa_outputpath	= NULL;
597 	args.pa_pplugin		= PMCSTAT_PL_NONE;
598 	args.pa_plugin		= PMCSTAT_PL_NONE;
599 	args.pa_ctdumpinstr	= 1;
600 	args.pa_topmode		= PMCSTAT_TOP_DELTA;
601 	args.pa_toptty		= 0;
602 	args.pa_topcolor	= 0;
603 	args.pa_mergepmc	= 0;
604 	args.pa_duration	= 0.0;
605 	STAILQ_INIT(&args.pa_events);
606 	SLIST_INIT(&args.pa_targets);
607 	bzero(&ds_start, sizeof(ds_start));
608 	bzero(&ds_end, sizeof(ds_end));
609 	ev = NULL;
610 	CPU_ZERO(&cpumask);
611 
612 	/* Default to using the running system kernel. */
613 	len = 0;
614 	if (sysctlbyname("kern.bootfile", NULL, &len, NULL, 0) == -1)
615 		err(EX_OSERR, "ERROR: Cannot determine path of running kernel");
616 	args.pa_kernel = malloc(len + 1);
617 	if (sysctlbyname("kern.bootfile", args.pa_kernel, &len, NULL, 0) == -1)
618 		err(EX_OSERR, "ERROR: Cannot determine path of running kernel");
619 
620 	/*
621 	 * The initial CPU mask specifies all non-halted CPUS in the
622 	 * system.
623 	 */
624 	len = sizeof(int);
625 	if (sysctlbyname("hw.ncpu", &ncpu, &len, NULL, 0) < 0)
626 		err(EX_OSERR, "ERROR: Cannot determine the number of CPUs");
627 	for (hcpu = 0; hcpu < ncpu; hcpu++)
628 		CPU_SET(hcpu, &cpumask);
629 
630 	while ((option = getopt(argc, argv,
631 	    "CD:EF:G:M:NO:P:R:S:TWa:c:df:gk:l:m:n:o:p:qr:s:t:vw:z:")) != -1)
632 		switch (option) {
633 		case 'a':	/* Annotate + callgraph */
634 			args.pa_flags |= FLAG_DO_ANNOTATE;
635 			args.pa_plugin = PMCSTAT_PL_ANNOTATE_CG;
636 			graphfilename  = optarg;
637 			break;
638 
639 		case 'C':	/* cumulative values */
640 			use_cumulative_counts = !use_cumulative_counts;
641 			args.pa_required |= FLAG_HAS_COUNTING_PMCS;
642 			break;
643 
644 		case 'c':	/* CPU */
645 
646 			if (optarg[0] == '*' && optarg[1] == '\0') {
647 				for (hcpu = 0; hcpu < ncpu; hcpu++)
648 					CPU_SET(hcpu, &cpumask);
649 			} else
650 				pmcstat_get_cpumask(optarg, &cpumask);
651 
652 			args.pa_flags	 |= FLAGS_HAS_CPUMASK;
653 			args.pa_required |= FLAG_HAS_SYSTEM_PMCS;
654 			break;
655 
656 		case 'D':
657 			if (stat(optarg, &sb) < 0)
658 				err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
659 				    optarg);
660 			if (!S_ISDIR(sb.st_mode))
661 				errx(EX_USAGE,
662 				    "ERROR: \"%s\" is not a directory.",
663 				    optarg);
664 			args.pa_samplesdir = optarg;
665 			args.pa_flags     |= FLAG_HAS_SAMPLESDIR;
666 			args.pa_required  |= FLAG_DO_GPROF;
667 			break;
668 
669 		case 'd':	/* toggle descendents */
670 			do_descendants = !do_descendants;
671 			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
672 			break;
673 
674 		case 'F':	/* produce a system-wide calltree */
675 			args.pa_flags |= FLAG_DO_CALLGRAPHS;
676 			args.pa_plugin = PMCSTAT_PL_CALLTREE;
677 			graphfilename = optarg;
678 			break;
679 
680 		case 'f':	/* plugins options */
681 			if (args.pa_plugin == PMCSTAT_PL_NONE)
682 				err(EX_USAGE, "ERROR: Need -g/-G/-m/-T.");
683 			pmcstat_pluginconfigure_log(optarg);
684 			break;
685 
686 		case 'G':	/* produce a system-wide callgraph */
687 			args.pa_flags |= FLAG_DO_CALLGRAPHS;
688 			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
689 			graphfilename = optarg;
690 			break;
691 
692 		case 'g':	/* produce gprof compatible profiles */
693 			args.pa_flags |= FLAG_DO_GPROF;
694 			args.pa_pplugin = PMCSTAT_PL_CALLGRAPH;
695 			args.pa_plugin	= PMCSTAT_PL_GPROF;
696 			break;
697 
698 		case 'k':	/* pathname to the kernel */
699 			free(args.pa_kernel);
700 			args.pa_kernel = strdup(optarg);
701 			args.pa_required |= FLAG_DO_ANALYSIS;
702 			args.pa_flags    |= FLAG_HAS_KERNELPATH;
703 			break;
704 
705 		case 'l':	/* time duration in seconds */
706 			duration = strtod(optarg, &end);
707 			if (*end != '\0' || duration <= 0)
708 				errx(EX_USAGE, "ERROR: Illegal duration time "
709 				    "value \"%s\".", optarg);
710 			args.pa_flags |= FLAG_HAS_DURATION;
711 			args.pa_duration = duration;
712 			break;
713 
714 		case 'm':
715 			args.pa_flags |= FLAG_DO_ANNOTATE;
716 			args.pa_plugin = PMCSTAT_PL_ANNOTATE;
717 			graphfilename  = optarg;
718 			break;
719 
720 		case 'E':	/* log process exit */
721 			do_logprocexit = !do_logprocexit;
722 			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
723 			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
724 			break;
725 
726 		case 'M':	/* mapfile */
727 			args.pa_mapfilename = optarg;
728 			break;
729 
730 		case 'N':
731 			do_callchain = !do_callchain;
732 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
733 			break;
734 
735 		case 'p':	/* process virtual counting PMC */
736 		case 's':	/* system-wide counting PMC */
737 		case 'P':	/* process virtual sampling PMC */
738 		case 'S':	/* system-wide sampling PMC */
739 			if ((ev = malloc(sizeof(*ev))) == NULL)
740 				errx(EX_SOFTWARE, "ERROR: Out of memory.");
741 
742 			switch (option) {
743 			case 'p': ev->ev_mode = PMC_MODE_TC; break;
744 			case 's': ev->ev_mode = PMC_MODE_SC; break;
745 			case 'P': ev->ev_mode = PMC_MODE_TS; break;
746 			case 'S': ev->ev_mode = PMC_MODE_SS; break;
747 			}
748 
749 			if (option == 'P' || option == 'p') {
750 				args.pa_flags |= FLAG_HAS_PROCESS_PMCS;
751 				args.pa_required |= (FLAG_HAS_COMMANDLINE |
752 				    FLAG_HAS_TARGET);
753 			}
754 
755 			if (option == 'P' || option == 'S') {
756 				args.pa_flags |= FLAG_HAS_SAMPLING_PMCS;
757 				args.pa_required |= (FLAG_HAS_PIPE |
758 				    FLAG_HAS_OUTPUT_LOGFILE);
759 			}
760 
761 			if (option == 'p' || option == 's')
762 				args.pa_flags |= FLAG_HAS_COUNTING_PMCS;
763 
764 			if (option == 's' || option == 'S')
765 				args.pa_flags |= FLAG_HAS_SYSTEM_PMCS;
766 
767 			ev->ev_spec  = strdup(optarg);
768 
769 			if (option == 'S' || option == 'P')
770 				ev->ev_count = current_sampling_count;
771 			else
772 				ev->ev_count = -1;
773 
774 			if (option == 'S' || option == 's') {
775 				hcpu = sizeof(cpumask) * NBBY;
776 				for (hcpu--; hcpu >= 0; hcpu--)
777 					if (CPU_ISSET(hcpu, &cpumask))
778 						break;
779 				ev->ev_cpu = hcpu;
780 			} else
781 				ev->ev_cpu = PMC_CPU_ANY;
782 
783 			ev->ev_flags = 0;
784 			if (do_callchain)
785 				ev->ev_flags |= PMC_F_CALLCHAIN;
786 			if (do_descendants)
787 				ev->ev_flags |= PMC_F_DESCENDANTS;
788 			if (do_logprocexit)
789 				ev->ev_flags |= PMC_F_LOG_PROCEXIT;
790 			if (do_logproccsw)
791 				ev->ev_flags |= PMC_F_LOG_PROCCSW;
792 
793 			ev->ev_cumulative  = use_cumulative_counts;
794 
795 			ev->ev_saved = 0LL;
796 			ev->ev_pmcid = PMC_ID_INVALID;
797 
798 			/* extract event name */
799 			c = strcspn(optarg, ", \t");
800 			ev->ev_name = malloc(c + 1);
801 			(void) strncpy(ev->ev_name, optarg, c);
802 			*(ev->ev_name + c) = '\0';
803 
804 			STAILQ_INSERT_TAIL(&args.pa_events, ev, ev_next);
805 
806 			if (option == 's' || option == 'S') {
807 				hcpu = CPU_ISSET(ev->ev_cpu, &cpumask);
808 				CPU_CLR(ev->ev_cpu, &cpumask);
809 				pmcstat_clone_event_descriptor(ev, &cpumask);
810 				if (hcpu != 0)
811 					CPU_SET(ev->ev_cpu, &cpumask);
812 			}
813 
814 			break;
815 
816 		case 'n':	/* sampling count */
817 			current_sampling_count = strtol(optarg, &end, 0);
818 			if (*end != '\0' || current_sampling_count <= 0)
819 				errx(EX_USAGE,
820 				    "ERROR: Illegal count value \"%s\".",
821 				    optarg);
822 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
823 			break;
824 
825 		case 'o':	/* outputfile */
826 			if (args.pa_printfile != NULL &&
827 			    args.pa_printfile != stdout &&
828 			    args.pa_printfile != stderr)
829 				(void) fclose(args.pa_printfile);
830 			if ((args.pa_printfile = fopen(optarg, "w")) == NULL)
831 				errx(EX_OSERR,
832 				    "ERROR: cannot open \"%s\" for writing.",
833 				    optarg);
834 			args.pa_flags |= FLAG_DO_PRINT;
835 			break;
836 
837 		case 'O':	/* sampling output */
838 			if (args.pa_outputpath)
839 				errx(EX_USAGE,
840 "ERROR: option -O may only be specified once.");
841 			args.pa_outputpath = optarg;
842 			args.pa_flags |= FLAG_HAS_OUTPUT_LOGFILE;
843 			break;
844 
845 		case 'q':	/* quiet mode */
846 			args.pa_verbosity = 0;
847 			break;
848 
849 		case 'r':	/* root FS path */
850 			args.pa_fsroot = optarg;
851 			break;
852 
853 		case 'R':	/* read an existing log file */
854 			if (args.pa_inputpath != NULL)
855 				errx(EX_USAGE,
856 "ERROR: option -R may only be specified once.");
857 			args.pa_inputpath = optarg;
858 			if (args.pa_printfile == stderr)
859 				args.pa_printfile = stdout;
860 			args.pa_flags |= FLAG_READ_LOGFILE;
861 			break;
862 
863 		case 't':	/* target pid or process name */
864 			pmcstat_find_targets(optarg);
865 
866 			args.pa_flags |= FLAG_HAS_TARGET;
867 			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
868 			break;
869 
870 		case 'T':	/* top mode */
871 			args.pa_flags |= FLAG_DO_TOP;
872 			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
873 			args.pa_ctdumpinstr = 0;
874 			args.pa_mergepmc = 1;
875 			if (args.pa_printfile == stderr)
876 				args.pa_printfile = stdout;
877 			break;
878 
879 		case 'v':	/* verbose */
880 			args.pa_verbosity++;
881 			break;
882 
883 		case 'w':	/* wait interval */
884 			interval = strtod(optarg, &end);
885 			if (*end != '\0' || interval <= 0)
886 				errx(EX_USAGE,
887 "ERROR: Illegal wait interval value \"%s\".",
888 				    optarg);
889 			args.pa_flags |= FLAG_HAS_WAIT_INTERVAL;
890 			args.pa_interval = interval;
891 			break;
892 
893 		case 'W':	/* toggle LOG_CSW */
894 			do_logproccsw = !do_logproccsw;
895 			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
896 			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
897 			break;
898 
899 		case 'z':
900 			graphdepth = strtod(optarg, &end);
901 			if (*end != '\0' || graphdepth <= 0)
902 				errx(EX_USAGE,
903 				    "ERROR: Illegal callchain depth \"%s\".",
904 				    optarg);
905 			args.pa_graphdepth = graphdepth;
906 			args.pa_required |= FLAG_DO_CALLGRAPHS;
907 			break;
908 
909 		case '?':
910 		default:
911 			pmcstat_show_usage();
912 			break;
913 
914 		}
915 
916 	args.pa_argc = (argc -= optind);
917 	args.pa_argv = (argv += optind);
918 
919 	/* If we read from logfile and no specified CPU mask use
920 	 * the maximum CPU count.
921 	 */
922 	if ((args.pa_flags & FLAG_READ_LOGFILE) &&
923 	    (args.pa_flags & FLAGS_HAS_CPUMASK) == 0)
924 		CPU_FILL(&cpumask);
925 
926 	args.pa_cpumask = cpumask; /* For selecting CPUs using -R. */
927 
928 	if (argc)	/* command line present */
929 		args.pa_flags |= FLAG_HAS_COMMANDLINE;
930 
931 	if (args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS |
932 	    FLAG_DO_ANNOTATE | FLAG_DO_TOP))
933 		args.pa_flags |= FLAG_DO_ANALYSIS;
934 
935 	/*
936 	 * Check invocation syntax.
937 	 */
938 
939 	/* disallow -O and -R together */
940 	if (args.pa_outputpath && args.pa_inputpath)
941 		errx(EX_USAGE,
942 		    "ERROR: options -O and -R are mutually exclusive.");
943 
944 	/* disallow -T and -l together */
945 	if ((args.pa_flags & FLAG_HAS_DURATION) &&
946 	    (args.pa_flags & FLAG_DO_TOP))
947 		errx(EX_USAGE, "ERROR: options -T and -l are mutually "
948 		    "exclusive.");
949 
950 	/* -a and -m require -R */
951 	if (args.pa_flags & FLAG_DO_ANNOTATE && args.pa_inputpath == NULL)
952 		errx(EX_USAGE, "ERROR: option %s requires an input file",
953 		    args.pa_plugin == PMCSTAT_PL_ANNOTATE ? "-m" : "-a");
954 
955 	/* -m option is not allowed combined with -g or -G. */
956 	if (args.pa_flags & FLAG_DO_ANNOTATE &&
957 	    args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS))
958 		errx(EX_USAGE,
959 		    "ERROR: option -m and -g | -G are mutually exclusive");
960 
961 	if (args.pa_flags & FLAG_READ_LOGFILE) {
962 		errmsg = NULL;
963 		if (args.pa_flags & FLAG_HAS_COMMANDLINE)
964 			errmsg = "a command line specification";
965 		else if (args.pa_flags & FLAG_HAS_TARGET)
966 			errmsg = "option -t";
967 		else if (!STAILQ_EMPTY(&args.pa_events))
968 			errmsg = "a PMC event specification";
969 		if (errmsg)
970 			errx(EX_USAGE,
971 			    "ERROR: option -R may not be used with %s.",
972 			    errmsg);
973 	} else if (STAILQ_EMPTY(&args.pa_events))
974 		/* All other uses require a PMC spec. */
975 		pmcstat_show_usage();
976 
977 	/* check for -t pid without a process PMC spec */
978 	if ((args.pa_required & FLAG_HAS_TARGET) &&
979 	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
980 		errx(EX_USAGE,
981 "ERROR: option -t requires a process mode PMC to be specified."
982 		    );
983 
984 	/* check for process-mode options without a command or -t pid */
985 	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
986 	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
987 		errx(EX_USAGE,
988 "ERROR: options -d, -E, -p, -P, and -W require a command line or target process."
989 		    );
990 
991 	/* check for -p | -P without a target process of some sort */
992 	if ((args.pa_required & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) &&
993 	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
994 		errx(EX_USAGE,
995 "ERROR: options -P and -p require a target process or a command line."
996 		    );
997 
998 	/* check for process-mode options without a process-mode PMC */
999 	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
1000 	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
1001 		errx(EX_USAGE,
1002 "ERROR: options -d, -E, and -W require a process mode PMC to be specified."
1003 		    );
1004 
1005 	/* check for -c cpu with no system mode PMCs or logfile. */
1006 	if ((args.pa_required & FLAG_HAS_SYSTEM_PMCS) &&
1007 	    (args.pa_flags & FLAG_HAS_SYSTEM_PMCS) == 0 &&
1008 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
1009 		errx(EX_USAGE,
1010 "ERROR: option -c requires at least one system mode PMC to be specified."
1011 		    );
1012 
1013 	/* check for counting mode options without a counting PMC */
1014 	if ((args.pa_required & FLAG_HAS_COUNTING_PMCS) &&
1015 	    (args.pa_flags & FLAG_HAS_COUNTING_PMCS) == 0)
1016 		errx(EX_USAGE,
1017 "ERROR: options -C, -W and -o require at least one counting mode PMC to be specified."
1018 		    );
1019 
1020 	/* check for sampling mode options without a sampling PMC spec */
1021 	if ((args.pa_required & FLAG_HAS_SAMPLING_PMCS) &&
1022 	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) == 0)
1023 		errx(EX_USAGE,
1024 "ERROR: options -N, -n and -O require at least one sampling mode PMC to be specified."
1025 		    );
1026 
1027 	/* check if -g/-G/-m/-T are being used correctly */
1028 	if ((args.pa_flags & FLAG_DO_ANALYSIS) &&
1029 	    !(args.pa_flags & (FLAG_HAS_SAMPLING_PMCS|FLAG_READ_LOGFILE)))
1030 		errx(EX_USAGE,
1031 "ERROR: options -g/-G/-m/-T require sampling PMCs or -R to be specified."
1032 		    );
1033 
1034 	/* check if -O was spuriously specified */
1035 	if ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) &&
1036 	    (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0)
1037 		errx(EX_USAGE,
1038 "ERROR: option -O is used only with options -E, -P, -S and -W."
1039 		    );
1040 
1041 	/* -k kernel path require -g/-G/-m/-T or -R */
1042 	if ((args.pa_flags & FLAG_HAS_KERNELPATH) &&
1043 	    (args.pa_flags & FLAG_DO_ANALYSIS) == 0 &&
1044 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
1045 	    errx(EX_USAGE, "ERROR: option -k is only used with -g/-R/-m/-T.");
1046 
1047 	/* -D only applies to gprof output mode (-g) */
1048 	if ((args.pa_flags & FLAG_HAS_SAMPLESDIR) &&
1049 	    (args.pa_flags & FLAG_DO_GPROF) == 0)
1050 	    errx(EX_USAGE, "ERROR: option -D is only used with -g.");
1051 
1052 	/* -M mapfile requires -g or -R */
1053 	if (args.pa_mapfilename != NULL &&
1054 	    (args.pa_flags & FLAG_DO_GPROF) == 0 &&
1055 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
1056 	    errx(EX_USAGE, "ERROR: option -M is only used with -g/-R.");
1057 
1058 	/*
1059 	 * Disallow textual output of sampling PMCs if counting PMCs
1060 	 * have also been asked for, mostly because the combined output
1061 	 * is difficult to make sense of.
1062 	 */
1063 	if ((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
1064 	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) &&
1065 	    ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) == 0))
1066 		errx(EX_USAGE,
1067 "ERROR: option -O is required if counting and sampling PMCs are specified together."
1068 		    );
1069 
1070 	/*
1071 	 * Check if 'kerneldir' refers to a file rather than a
1072 	 * directory.  If so, use `dirname path` to determine the
1073 	 * kernel directory.
1074 	 */
1075 	(void) snprintf(buffer, sizeof(buffer), "%s%s", args.pa_fsroot,
1076 	    args.pa_kernel);
1077 	if (stat(buffer, &sb) < 0)
1078 		err(EX_OSERR, "ERROR: Cannot locate kernel \"%s\"",
1079 		    buffer);
1080 	if (!S_ISREG(sb.st_mode) && !S_ISDIR(sb.st_mode))
1081 		errx(EX_USAGE, "ERROR: \"%s\": Unsupported file type.",
1082 		    buffer);
1083 	if (!S_ISDIR(sb.st_mode)) {
1084 		tmp = args.pa_kernel;
1085 		args.pa_kernel = strdup(dirname(args.pa_kernel));
1086 		free(tmp);
1087 		(void) snprintf(buffer, sizeof(buffer), "%s%s",
1088 		    args.pa_fsroot, args.pa_kernel);
1089 		if (stat(buffer, &sb) < 0)
1090 			err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
1091 			    buffer);
1092 		if (!S_ISDIR(sb.st_mode))
1093 			errx(EX_USAGE,
1094 			    "ERROR: \"%s\" is not a directory.",
1095 			    buffer);
1096 	}
1097 
1098 	/*
1099 	 * If we have a callgraph be created, select the outputfile.
1100 	 */
1101 	if (args.pa_flags & FLAG_DO_CALLGRAPHS) {
1102 		if (strcmp(graphfilename, "-") == 0)
1103 		    args.pa_graphfile = args.pa_printfile;
1104 		else {
1105 			args.pa_graphfile = fopen(graphfilename, "w");
1106 			if (args.pa_graphfile == NULL)
1107 				err(EX_OSERR,
1108 				    "ERROR: cannot open \"%s\" for writing",
1109 				    graphfilename);
1110 		}
1111 	}
1112 	if (args.pa_flags & FLAG_DO_ANNOTATE) {
1113 		args.pa_graphfile = fopen(graphfilename, "w");
1114 		if (args.pa_graphfile == NULL)
1115 			err(EX_OSERR, "ERROR: cannot open \"%s\" for writing",
1116 			    graphfilename);
1117 	}
1118 
1119 	/* if we've been asked to process a log file, skip init */
1120 	if ((args.pa_flags & FLAG_READ_LOGFILE) == 0) {
1121 		if (pmc_init() < 0)
1122 			err(EX_UNAVAILABLE,
1123 			    "ERROR: Initialization of the pmc(3) library failed"
1124 			    );
1125 
1126 		if ((npmc = pmc_npmc(0)) < 0) /* assume all CPUs are identical */
1127 			err(EX_OSERR,
1128 "ERROR: Cannot determine the number of PMCs on CPU %d",
1129 			    0);
1130 	}
1131 
1132 	/* Allocate a kqueue */
1133 	if ((pmcstat_kq = kqueue()) < 0)
1134 		err(EX_OSERR, "ERROR: Cannot allocate kqueue");
1135 
1136 	/* Setup the logfile as the source. */
1137 	if (args.pa_flags & FLAG_READ_LOGFILE) {
1138 		/*
1139 		 * Print the log in textual form if we haven't been
1140 		 * asked to generate profiling information.
1141 		 */
1142 		if ((args.pa_flags & FLAG_DO_ANALYSIS) == 0)
1143 			args.pa_flags |= FLAG_DO_PRINT;
1144 
1145 		pmcstat_initialize_logging();
1146 		rfd = pmcstat_open_log(args.pa_inputpath,
1147 		    PMCSTAT_OPEN_FOR_READ);
1148 		if ((args.pa_logparser = pmclog_open(rfd)) == NULL)
1149 			err(EX_OSERR, "ERROR: Cannot create parser");
1150 		if (fcntl(rfd, F_SETFL, O_NONBLOCK) < 0)
1151 			err(EX_OSERR, "ERROR: fcntl(2) failed");
1152 		EV_SET(&kev, rfd, EVFILT_READ, EV_ADD,
1153 		    0, 0, NULL);
1154 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1155 			err(EX_OSERR, "ERROR: Cannot register kevent");
1156 	}
1157 	/*
1158 	 * Configure the specified log file or setup a default log
1159 	 * consumer via a pipe.
1160 	 */
1161 	if (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) {
1162 		if (args.pa_outputpath)
1163 			args.pa_logfd = pmcstat_open_log(args.pa_outputpath,
1164 			    PMCSTAT_OPEN_FOR_WRITE);
1165 		else {
1166 			/*
1167 			 * process the log on the fly by reading it in
1168 			 * through a pipe.
1169 			 */
1170 			if (pipe(pipefd) < 0)
1171 				err(EX_OSERR, "ERROR: pipe(2) failed");
1172 
1173 			if (fcntl(pipefd[READPIPEFD], F_SETFL, O_NONBLOCK) < 0)
1174 				err(EX_OSERR, "ERROR: fcntl(2) failed");
1175 
1176 			EV_SET(&kev, pipefd[READPIPEFD], EVFILT_READ, EV_ADD,
1177 			    0, 0, NULL);
1178 
1179 			if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1180 				err(EX_OSERR, "ERROR: Cannot register kevent");
1181 
1182 			args.pa_logfd = pipefd[WRITEPIPEFD];
1183 
1184 			args.pa_flags |= FLAG_HAS_PIPE;
1185 			if ((args.pa_flags & FLAG_DO_TOP) == 0)
1186 				args.pa_flags |= FLAG_DO_PRINT;
1187 			args.pa_logparser = pmclog_open(pipefd[READPIPEFD]);
1188 		}
1189 
1190 		if (pmc_configure_logfile(args.pa_logfd) < 0)
1191 			err(EX_OSERR, "ERROR: Cannot configure log file");
1192 	}
1193 
1194 	/* remember to check for driver errors if we are sampling or logging */
1195 	check_driver_stats = (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) ||
1196 	    (args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE);
1197 
1198 	/*
1199 	if (args.pa_flags & FLAG_READ_LOGFILE) {
1200 	 * Allocate PMCs.
1201 	 */
1202 
1203 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1204 		if (pmc_allocate(ev->ev_spec, ev->ev_mode,
1205 		    ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid) < 0)
1206 			err(EX_OSERR,
1207 "ERROR: Cannot allocate %s-mode pmc with specification \"%s\"",
1208 			    PMC_IS_SYSTEM_MODE(ev->ev_mode) ?
1209 			    "system" : "process", ev->ev_spec);
1210 
1211 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode) &&
1212 		    pmc_set(ev->ev_pmcid, ev->ev_count) < 0)
1213 			err(EX_OSERR,
1214 			    "ERROR: Cannot set sampling count for PMC \"%s\"",
1215 			    ev->ev_name);
1216 	}
1217 
1218 	/* compute printout widths */
1219 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1220 		int counter_width;
1221 		int display_width;
1222 		int header_width;
1223 
1224 		(void) pmc_width(ev->ev_pmcid, &counter_width);
1225 		header_width = strlen(ev->ev_name) + 2; /* prefix '%c/' */
1226 		display_width = (int) floor(counter_width / 3.32193) + 1;
1227 
1228 		if (PMC_IS_SYSTEM_MODE(ev->ev_mode))
1229 			header_width += 3; /* 2 digit CPU number + '/' */
1230 
1231 		if (header_width > display_width) {
1232 			ev->ev_fieldskip = 0;
1233 			ev->ev_fieldwidth = header_width;
1234 		} else {
1235 			ev->ev_fieldskip = display_width -
1236 			    header_width;
1237 			ev->ev_fieldwidth = display_width;
1238 		}
1239 	}
1240 
1241 	/*
1242 	 * If our output is being set to a terminal, register a handler
1243 	 * for window size changes.
1244 	 */
1245 
1246 	if (isatty(fileno(args.pa_printfile))) {
1247 
1248 		if (ioctl(fileno(args.pa_printfile), TIOCGWINSZ, &ws) < 0)
1249 			err(EX_OSERR, "ERROR: Cannot determine window size");
1250 
1251 		pmcstat_displayheight = ws.ws_row - 1;
1252 		pmcstat_displaywidth  = ws.ws_col - 1;
1253 
1254 		EV_SET(&kev, SIGWINCH, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1255 
1256 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1257 			err(EX_OSERR,
1258 			    "ERROR: Cannot register kevent for SIGWINCH");
1259 
1260 		args.pa_toptty = 1;
1261 	}
1262 
1263 	/*
1264 	 * Listen to key input in top mode.
1265 	 */
1266 	if (args.pa_flags & FLAG_DO_TOP) {
1267 		EV_SET(&kev, fileno(stdin), EVFILT_READ, EV_ADD, 0, 0, NULL);
1268 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1269 			err(EX_OSERR, "ERROR: Cannot register kevent");
1270 	}
1271 
1272 	EV_SET(&kev, SIGINT, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1273 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1274 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGINT");
1275 
1276 	EV_SET(&kev, SIGIO, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1277 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1278 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGIO");
1279 
1280 	/*
1281 	 * An exec() failure of a forked child is signalled by the
1282 	 * child sending the parent a SIGCHLD.  We don't register an
1283 	 * actual signal handler for SIGCHLD, but instead use our
1284 	 * kqueue to pick up the signal.
1285 	 */
1286 	EV_SET(&kev, SIGCHLD, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1287 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1288 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGCHLD");
1289 
1290 	/*
1291 	 * Setup a timer if we have counting mode PMCs needing to be printed or
1292 	 * top mode plugin is active.
1293 	 */
1294 	if (((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
1295 	     (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) ||
1296 	    (args.pa_flags & FLAG_DO_TOP)) {
1297 		EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0,
1298 		    args.pa_interval * 1000, NULL);
1299 
1300 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1301 			err(EX_OSERR,
1302 			    "ERROR: Cannot register kevent for timer");
1303 	}
1304 
1305 	/*
1306 	 * Setup a duration timer if we have sampling mode PMCs and
1307 	 * a duration time is set
1308 	 */
1309 	if ((args.pa_flags & FLAG_HAS_SAMPLING_PMCS) &&
1310 	    (args.pa_flags & FLAG_HAS_DURATION)) {
1311 		EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0,
1312 		    args.pa_duration * 1000, NULL);
1313 
1314 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1315 			err(EX_OSERR, "ERROR: Cannot register kevent for "
1316 			    "time duration");
1317 	}
1318 
1319 	/* attach PMCs to the target process, starting it if specified */
1320 	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1321 		pmcstat_create_process();
1322 
1323 	if (check_driver_stats && pmc_get_driver_stats(&ds_start) < 0)
1324 		err(EX_OSERR, "ERROR: Cannot retrieve driver statistics");
1325 
1326 	/* Attach process pmcs to the target process. */
1327 	if (args.pa_flags & (FLAG_HAS_TARGET | FLAG_HAS_COMMANDLINE)) {
1328 		if (SLIST_EMPTY(&args.pa_targets))
1329 			errx(EX_DATAERR,
1330 			    "ERROR: No matching target processes.");
1331 		if (args.pa_flags & FLAG_HAS_PROCESS_PMCS)
1332 			pmcstat_attach_pmcs();
1333 
1334 		if (pmcstat_kvm) {
1335 			kvm_close(pmcstat_kvm);
1336 			pmcstat_kvm = NULL;
1337 		}
1338 	}
1339 
1340 	/* start the pmcs */
1341 	pmcstat_start_pmcs();
1342 
1343 	/* start the (commandline) process if needed */
1344 	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1345 		pmcstat_start_process();
1346 
1347 	/* initialize logging */
1348 	pmcstat_initialize_logging();
1349 
1350 	/* Handle SIGINT using the kqueue loop */
1351 	sa.sa_handler = SIG_IGN;
1352 	sa.sa_flags   = 0;
1353 	(void) sigemptyset(&sa.sa_mask);
1354 
1355 	if (sigaction(SIGINT, &sa, NULL) < 0)
1356 		err(EX_OSERR, "ERROR: Cannot install signal handler");
1357 
1358 	/*
1359 	 * Setup the top mode display.
1360 	 */
1361 	if (args.pa_flags & FLAG_DO_TOP) {
1362 		args.pa_flags &= ~FLAG_DO_PRINT;
1363 
1364 		if (args.pa_toptty) {
1365 			/*
1366 			 * Init ncurses.
1367 			 */
1368 			initscr();
1369 			if(has_colors() == TRUE) {
1370 				args.pa_topcolor = 1;
1371 				start_color();
1372 				use_default_colors();
1373 				pair_content(0, &cf, &cb);
1374 				init_pair(1, COLOR_RED, cb);
1375 				init_pair(2, COLOR_YELLOW, cb);
1376 				init_pair(3, COLOR_GREEN, cb);
1377 			}
1378 			cbreak();
1379 			noecho();
1380 			nonl();
1381 			nodelay(stdscr, 1);
1382 			intrflush(stdscr, FALSE);
1383 			keypad(stdscr, TRUE);
1384 			clear();
1385 			/* Get terminal width / height with ncurses. */
1386 			getmaxyx(stdscr,
1387 			    pmcstat_displayheight, pmcstat_displaywidth);
1388 			pmcstat_displayheight--; pmcstat_displaywidth--;
1389 			atexit(pmcstat_topexit);
1390 		}
1391 	}
1392 
1393 	/*
1394 	 * loop till either the target process (if any) exits, or we
1395 	 * are killed by a SIGINT or we reached the time duration.
1396 	 */
1397 	runstate = PMCSTAT_RUNNING;
1398 	do_print = do_read = 0;
1399 	do {
1400 		if ((c = kevent(pmcstat_kq, NULL, 0, &kev, 1, NULL)) <= 0) {
1401 			if (errno != EINTR)
1402 				err(EX_OSERR, "ERROR: kevent failed");
1403 			else
1404 				continue;
1405 		}
1406 
1407 		if (kev.flags & EV_ERROR)
1408 			errc(EX_OSERR, kev.data, "ERROR: kevent failed");
1409 
1410 		switch (kev.filter) {
1411 		case EVFILT_PROC:  /* target has exited */
1412 			runstate = pmcstat_close_log();
1413 			do_print = 1;
1414 			break;
1415 
1416 		case EVFILT_READ:  /* log file data is present */
1417 			if (kev.ident == (unsigned)fileno(stdin) &&
1418 			    (args.pa_flags & FLAG_DO_TOP)) {
1419 				if (pmcstat_keypress_log())
1420 					runstate = pmcstat_close_log();
1421 			} else {
1422 				do_read = 0;
1423 				runstate = pmcstat_process_log();
1424 			}
1425 			break;
1426 
1427 		case EVFILT_SIGNAL:
1428 			if (kev.ident == SIGCHLD) {
1429 				/*
1430 				 * The child process sends us a
1431 				 * SIGCHLD if its exec() failed.  We
1432 				 * wait for it to exit and then exit
1433 				 * ourselves.
1434 				 */
1435 				(void) wait(&c);
1436 				runstate = PMCSTAT_FINISHED;
1437 			} else if (kev.ident == SIGIO) {
1438 				/*
1439 				 * We get a SIGIO if a PMC loses all
1440 				 * of its targets, or if logfile
1441 				 * writes encounter an error.
1442 				 */
1443 				runstate = pmcstat_close_log();
1444 				do_print = 1; /* print PMCs at exit */
1445 			} else if (kev.ident == SIGINT) {
1446 				/* Kill the child process if we started it */
1447 				if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1448 					pmcstat_kill_process();
1449 				runstate = pmcstat_close_log();
1450 			} else if (kev.ident == SIGWINCH) {
1451 				if (ioctl(fileno(args.pa_printfile),
1452 					TIOCGWINSZ, &ws) < 0)
1453 				    err(EX_OSERR,
1454 				        "ERROR: Cannot determine window size");
1455 				pmcstat_displayheight = ws.ws_row - 1;
1456 				pmcstat_displaywidth  = ws.ws_col - 1;
1457 			} else
1458 				assert(0);
1459 
1460 			break;
1461 
1462 		case EVFILT_TIMER:
1463 			/* time duration reached, exit */
1464 			if (args.pa_flags & FLAG_HAS_DURATION) {
1465 				runstate = PMCSTAT_FINISHED;
1466 				break;
1467 			}
1468 			/* print out counting PMCs */
1469 			if ((args.pa_flags & FLAG_DO_TOP) &&
1470 			     pmc_flush_logfile() == 0)
1471 				do_read = 1;
1472 			do_print = 1;
1473 			break;
1474 
1475 		}
1476 
1477 		if (do_print && !do_read) {
1478 			if ((args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) {
1479 				pmcstat_print_pmcs();
1480 				if (runstate == PMCSTAT_FINISHED &&
1481 				    /* final newline */
1482 				    (args.pa_flags & FLAG_DO_PRINT) == 0)
1483 					(void) fprintf(args.pa_printfile, "\n");
1484 			}
1485 			if (args.pa_flags & FLAG_DO_TOP)
1486 				pmcstat_display_log();
1487 			do_print = 0;
1488 		}
1489 
1490 	} while (runstate != PMCSTAT_FINISHED);
1491 
1492 	if ((args.pa_flags & FLAG_DO_TOP) && args.pa_toptty) {
1493 		pmcstat_topexit();
1494 		args.pa_toptty = 0;
1495 	}
1496 
1497 	/* flush any pending log entries */
1498 	if (args.pa_flags & (FLAG_HAS_OUTPUT_LOGFILE | FLAG_HAS_PIPE))
1499 		pmc_close_logfile();
1500 
1501 	pmcstat_cleanup();
1502 
1503 	free(args.pa_kernel);
1504 
1505 	/* check if the driver lost any samples or events */
1506 	if (check_driver_stats) {
1507 		if (pmc_get_driver_stats(&ds_end) < 0)
1508 			err(EX_OSERR,
1509 			    "ERROR: Cannot retrieve driver statistics");
1510 		if (ds_start.pm_intr_bufferfull != ds_end.pm_intr_bufferfull &&
1511 		    args.pa_verbosity > 0)
1512 			warnx("WARNING: some samples were dropped.\n"
1513 "Please consider tuning the \"kern.hwpmc.nsamples\" tunable."
1514 			    );
1515 		if (ds_start.pm_buffer_requests_failed !=
1516 		    ds_end.pm_buffer_requests_failed &&
1517 		    args.pa_verbosity > 0)
1518 			warnx("WARNING: some events were discarded.\n"
1519 "Please consider tuning the \"kern.hwpmc.nbuffers\" tunable."
1520 			    );
1521 	}
1522 
1523 	exit(EX_OK);
1524 }
1525